On 19 August several vetted cybersecurity researchers found their access to OpenAI's Trusted Access for Cyber programme — specifically the Daybreak Blue tier — had been revoked. The Cyber page told them their identity "could not be verified" or that their account "is ineligible at this time." TechCrunch reported it at 18:46 UTC.
What OpenAI says
The company's position is that this was accidental. In an email it said: "This was an issue on our end, and not the user experience we want to deliver." Its forum support cited a "recent technical issue," and a post stated that "a limited set of users' access to Daybreak Blue is no longer active and they will need to re-verify." Affected researchers were asked to reapply and re-verify rather than appeal.
The pattern in the affected set
All five researchers who spoke to TechCrunch live outside the US and Europe. That is the only common thread anyone has identified, and it is exactly the failure mode you would expect from an automated identity-verification system: not a policy decision about who deserves access, but a verification pipeline that works less well the further you get from its training ground.
What the common telling gets wrong
Three corrections. "OpenAI bans researchers" imputes a policy motive the company denies and the evidence does not support. It is also not a shutdown of the programme — Daybreak Blue stayed live while a subset of already-vetted accounts lost eligibility. And the shorthand "reduced guardrails" misstates what the tier is: OpenAI describes it as access to frontier general-purpose models including GPT-5.6 Sol "with safeguards tailored to authorized defensive security work," which is a different thing from a model with its safety training removed.
The number nobody has
OpenAI has not disclosed how many accounts were affected. "A limited number of users" is the company's own characterisation of its own incident, and no independent count exists. We are not publishing a figure because none has been established.